No Reference Quality Assessment for Screen Content Images Based on Entire and High-Influence Regions

Screen content images (SCIs) are composed of text, graphics, natural scene images and other contents, for which the human eye will have different visual perception for different regions. Considering the impact of local distortions on the visual quality of the entire SCIs, this paper proposes a no-reference quality assessment method based on entire and high-influence regions. Firstly, we select the region with higher influence on the overall quality of the image by information entropy. Then the structural features based on phase congruency and color features based on opponent color space are extracted from the entire SCIs and high-influence region. The AdaBoosting neural network (ABNN) is adopted to predict scores of the entire image and the high-influence region. Finally, the quality score of the high-influence is used to slightly adjust the score of the entire SCIs to obtain the final score through the weighting strategy. Extensive experiments demonstrate the efficiency and robustness of the proposed method.